02 +
i Al Da
bap 5 ses
Bouzidi, Sonia
errors observed for the land covers grass and field are important. We are currently wondering about the kind of qualitative
labels to create. We also suppose a confusion between these two classes because of their similar visible and near infra-red
behaviors which have been used for the unmixing process of NOAA data. This is illustrated on figure 6 by displaying
the NDVI (Normalized Derived Vegetation Index) curves estimated from the visible and near infra-red NOAA/AVHRR
channels (N.DVI — (nir — vis)/(nir -- vis)): the curves are quite similar and that surely explains the difficulty to
discriminate grass and fields in the visible and near infra-red channels. A future task is then to compute a differentiation
Figure 6: Temporal profiles of NDVI, for field and grass.
criterion between these two land covers, in order to minimize this confusion. To do this, we propose to make use of the
thermal channels and to compute proportions by considering the physical based mixture model that has been discussed
in (Lahoche et al., 2000).
5 CONCLUSION
We describe a method for land use classification at meso-scale using coarse spatial resolution remotely sensed data ac-
quired by the NOAA/AVHRR sensors. We define a model for the pixels’ content and a process allowing to compute the
individual proportions of the land covers for each pixel. This process exploits the temporal information of the NOAA data
and is based on inverting a linear mixture model of reflectance, we call it the *unmixing process". The result provides a
description in terms of land covers percentage within each NOAA pixel. Results evaluation is carried out on a test site and
show that the method gives a global idea about land cover proportion inside the NOAA pixels. If this method seems to
be promising, it must still be improved: define which types of land cover can be discriminated, and with which precision
level. On another hand, for the inversion of the linear mixture model, we have considered a determinist resolution. A new
formulation based on a probabilistic framework is under investigation, in order to estimate proportion values and also the
associated likelihood.
6 ACKNOWLEDGMENTS
This research is made within the context of the European INCO-PED IWRMS project (Integrated Water Resources Man-
agement System http://www.iwrms.uni-jena.de/) funded by European Commission under the contract ERBIC18CT97044.
REFERENCES
Bouzidi, S., Berroir, J.-P. and Herlin, L, 1997a. Simultaneous use of SPOT and NOAA/AVHRR data for vegetation
monitoring. In: proceedings of the 10th Scandinavian Conference on Image Analysis.
Bouzidi, S., Berroir, J.-P. and Herlin, L, 1997b. Subpixel mixture modeling applied for vegetation monitoring. In:
proceedings of the International Symposium on Environmental Software Systems.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 211